Industries / Consumer goods / Sporting goods / Bicycles
Bike sharing
Bike sharing rents bicycles and e-bikes for single trips, from docking stations or from the street, in systems run by cities, advertising firms and private companies.
7
Milestones achieved, from 1965 to 2019.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1965
White bikes in Amsterdam
In summer 1965 Provo, a group of Dutch anarchist activists, painted a small number of used bicycles white and left them without locks around Amsterdam for anyone to ride, as a protest against cars and air pollution. The idea came from the industrial engineer Luud Schimmelpennink. The police soon removed the bikes, and Schimmelpennink later called the first plan "just a symbolic thing"; his fuller plan, for the city to spread 10,000 white bikes, was rejected unanimously by the city council. In a 2011 paper, Susan Shaheen and colleagues at the University of California, Berkeley, call such free bikes without locks the first generation of bike sharing and note that they were often damaged or stolen.
LessonBikes with no lock and no record of who took them do not stay in service.
Credit
Provo; Luud Schimmelpennink
1990s
A coin releases the bike
In the mid-1990s two Danes asked Schimmelpennink to help build a system in Copenhagen, which the Guardian calls the world's first large-scale bike-share programme. A rider put a coin in the bike to release it and got the coin back on returning the bike. "A lot of bikes got stolen though, because the users could remain anonymous," Schimmelpennink told the Guardian. Shaheen and colleagues class such coin-deposit systems as the second generation and add that, with no time limit, users often kept bikes for long periods. Amsterdam tried again in 1999 with 250 bikes at five stations, released with a bank's chip card at one guilder a trip; the scheme lost bikes to theft and ended when the bank, Postbank, withdrew the card.
LessonKnowing who has a bike, and for how long, mattered more than the lock.
Credit
Luud Schimmelpennink and the Copenhagen organisers
2008
Hangzhou builds a public fleet
On 1 May 2008 the Hangzhou Public Transport Corporation, a state-owned company, launched Hangzhou Public Bicycle with 2,800 bicycles, 30 fixed stations and 31 stations that could be moved until demand was known. The city put in 180 million yuan and lent 270 million more at a discount. Riders used the city's transit smart card, the first hour was free, and plain one-speed bikes kept the cost of theft and vandalism down. Shaheen and colleagues call it the first bike-share system in mainland China run on information technology, and counted 60,600 bicycles at 2,416 stations by March 2011. Guinness World Records listed it in September 2016 as the world's largest, with 84,100 bikes and 3,572 stations, against 23,600 bikes for Paris's Vélib'.
Credit
Hangzhou city government; Hangzhou Public Transport Corporation
2014
One feed for every system
The General Bikeshare Feed Specification (GBFS) is an open format in which a system publishes, in real time, where its stations are, how many bikes and free docks each has, and where free-floating vehicles stand, so that any journey-planning app can show it without a separate deal. It was created in 2014 by Mitch Vars with operators, cities and software firms; staff of Motivate, now part of Lyft, made large contributions to version 1.0. Its success came with the endorsement and hosting of the North American Bikeshare Association from 2015, and since 2019 MobilityData has run it. It leaves out anything that identifies a rider. The public catalogue of systems using GBFS listed 1,283 systems in 47 countries on 18 November 2026, by this file's count, including 277 in France, 191 in Germany and 177 in the United States; the count includes scooter, moped and car services.
Credit
Mitch Vars; Motivate; the North American Bikeshare Association; MobilityData
2018
Deposits that never came back
In China, bikes that a phone app releases and that can be left anywhere, without docks, were led by ofo and its rival Mobike, and ofo required each rider to pay a deposit, 99 yuan (about $14) for early users and 199 yuan for later ones. By December 2018 ofo had raised no money from investors since March, user screenshots showed more than 10 million people waiting for refunds, and hundreds queued at its Beijing headquarters. China's transport ministry urged it to speed up repayment, and its chief executive, Dai Wei, said the company did not "correctly assess the changing external environment".
LessonA fleet funded with riders' deposits and investors' money left riders unpaid when the money stopped.
Credit
ofo; reported by Caixin and Sixth Tone
2018
A handover empties Vélib'
Paris launched Vélib' in 2007 with JCDecaux, the street-advertising company, which had already set up systems in Vienna and Lyon. It became the world's largest system of its time, with 20,600 bicycles at 1,431 stations and an average of 108,000 uses a day. For 2018 the city gave a new contract to the Smovengo consortium, to reach more towns around Paris and add electric bikes. JCDecaux had to take down every station before the new ones could go in, and the bikes available to Vélib's 300,000 subscribers fell from 10,000 in October 2017 to 600 in January 2018, with daily trips under 6,000, by the political scientist Maxime Huré's account.
LessonChanging operator meant rebuilding the network on the street, and the city lost most of its service for months.
Credit
City of Paris; JCDecaux; Smovengo; analysis by Maxime Huré in Métropolitiques
2019
E-bikes pulled for brakes
In April 2019 Lyft took about 3,000 pedal-assist e-bikes out of service in New York's Citi Bike, Washington's Capital Bikeshare and the San Francisco Bay Area's Ford GoBike, out of about 20,000 bikes in those three systems. Riders had reported that the front brake was stronger than expected and could cause a fall. Lyft worked with its supplier and an outside engineering firm to find the cause and kept its classic bikes running.
LessonA fault in one fleet model takes thousands of bikes off the street at once.
Credit
Lyft; reported by NBC News
Watch for
Feeds checked for meaning
Rohan Fossé and Gaël Pallares published, in May 2026, an open pipeline and dataset that audit GBFS feeds for errors a format check cannot catch, such as car-sharing stations listed as bike stations, placeholder records and stations far outside a system's area. They say an audit of 1,509 systems in 48 countries yields seven classes of problem, and that their rules remove 30.9% of the raw French stations and relabel a further 61%. Their paper is described as in preparation.
Achieved whenthe method and its precision against human checks are published after peer review, or operators correct the records it flags.
Claimed by
Rohan Fossé and Gaël Pallares, May 2026
Open
01
How long shared bikes last
A public table and survival curves showing how long London's Santander Cycles bikes stay in service, how many trips and hours each gives, and how long each sits unused between trips, as a stand-in for time in the workshop, split by model (classic or e-bike) and by year of first use, from Transport for London's open journey data.
The briefCredit · unclaimed02
Feed against the street
A repeatable check of whether a system's live GBFS feed matches what is at the station. At each visit: a timed phone photo of the station, framed to leave out people; the bikes and e-bikes counted from it; each marked rideable or not (flat tyre, missing saddle, visible damage); and the counts compared with the feed at the same minute.
The briefCredit · unclaimed
Open 01 · the brief
How long shared bikes last
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table and survival curves showing how long London's Santander Cycles bikes stay in service, how many trips and hours each gives, and how long each sits unused between trips, as a stand-in for time in the workshop, split by model (classic or e-bike) and by year of first use, from Transport for London's open journey data.
Why it is still open
TfL's journey files still give a bike number and model for every hire, as checked for this file in the latest file, for late May 2026, and the archive runs back to 2012. TfL has given fleet totals, such as about 200 of its 11,500 bikes scrapped each year for damage and 10,782 broken pedals in a year, as road.cc reported in 2019, but no figures per bike. A GitHub analysis estimated Citi Bike lifespans from each bike's first and last ride, 5.8 years for the 2013 bikes and 3.5 years for the 2016 bikes, but Citi Bike stopped publishing bike IDs in early 2021. A 2024 study of Barcelona's Bicing predicted when bike parts would fail from trip and repair records, not how long whole bikes last. No such study for London was found.
Built with what exists
TfL's usage-stats files at cycling.data.tfl.gov.uk; Python with DuckDB or pandas; the lifelines package for survival analysis.
First step
Load one year of journeys into DuckDB and list, for each bike number, its first hire, last hire and longest gap.
Done looks like
Every bike number from 2012 to 2026 processed; median service life and trips per bike by model and year of first use; the share of days each bike was in use; the code published; and the fleet size the data implies compared with TfL's published fleet figures for at least three dates.
Open 02 · the brief
Feed against the street
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable check of whether a system's live GBFS feed matches what is at the station. At each visit: a timed phone photo of the station, framed to leave out people; the bikes and e-bikes counted from it; each marked rideable or not (flat tyre, missing saddle, visible damage); and the counts compared with the feed at the same minute.
Why it is still open
The checks found test the feed, not the street. MobilityData's validator checks that a feed conforms to the specification, and the 2026 audit by Fossé and Pallares checks station records against imagery and flags stations whose status never changes, but neither counts bikes on site. A web search and a GitHub search for GBFS accuracy, ground truth and audits found only these and studies that use feeds to infer trips.
Built with what exists
Public GBFS feeds; a Python script that saves each station's status every minute; a phone; an AI model such as Claude or Gemini to count bikes in each photo, checked by a person; Gemini Decision 1 to give a yes-or-no probability that each pictured bike is rideable (see the Note "A thousand questions, one pass").
First step
Pick 20 stations on one commuting route, save their feed, and photograph each station twice a day for a week.
Done looks like
At least 300 station visits in three cities, published with photos, feed snapshots and hand counts; the share of visits where the feed's count of available bikes matched the street; the average gap; and the share of bikes listed as available that were not rideable.